Updates to fiscal forecasting models

Our October 2015 Forecast evaluation report set out our approach for assessing and scrutinising the wide range of forecasting models we use to produce our detailed bottom-up fiscal forecasts. We judge the performance of these models against a number of criteria, including:

• accuracy – how well does the model match outturns?;
• plausibility – how well do the model outputs align with theory and experience?;
• transparency – how easily can the model outputs be understood and scrutinised?; and
• effectiveness – how well does the model capture the tax or benefit system?

In line with these criteria, we have reviewed the HMRC modelling of two of our largest receipts items and have made the changes described below.

VAT deductions are an adjustment made to the HMRC VTTL (VAT theoretical tax liability) model to remove items that are not scored as net VAT receipts by the ONS. In 2014, around 64 per cent of these deductions reflected current VAT refunds to central government. The forecast for these deductions was previously projected in line with past trends. But, thanks to the public spending cuts implemented in recent years, deductions relating to the government sector have not risen as quickly as past trends would have suggested, so we have been over-forecasting deductions (and thus under-forecasting net VAT receipts). This error has built up as the fiscal consolidation has continued. The model was insufficiently transparent to pick it up sooner.

To correct this error, we have adjusted these government deductions to grow in line with our forecasts for government spending. This correction has boosted VAT receipts by around £0.4 billion in 2015-16, rising to £3.3 billion in 2020-21.

Class 1 national insurance contributions (NIC1) have previously been forecast by HMRC using a model originally created by the Government Actuary’s Department (GAD). This model also lacked transparency, making it challenging to scrutinise forecast outputs effectively. To improve consistency across income taxes, we have now aligned the Class 1 NICs forecast model with that used by HMRC to forecast PAYE receipts for us. The new model features the latest ONS data on the income distribution and allows for changes in this over time. This switch has boosted the NIC1 forecast by £2.8 billion by 2020-21, of which £2.6 billion comes from employer NIC1 contributions.

Moving to this HMRC model has increased the transparency of the forecast, giving us more scope to scrutinise and implement key judgements, in particular regarding the average marginal tax rates that drive the receipts forecast. The £2.8 billion upward revision reflects a £3.3 billion boost from assuming that a higher (and more plausible) proportion of income is taxed above the upper earnings limit, partly offset by a £0.5 billion loss from assuming that a lower proportion of income is taxed between the primary threshold and the upper earnings limit.

We will continue to work with the relevant specialists in forecasting departments to improve our fiscal forecasting models in line with these criteria. We are grateful for the commitment, expertise and professionalism consistently shown by the analysts that work on these forecast models for us, often at times when Government calls on their time are significant.